## Does the BrainGate2 Speech BCI Work Unsupervised at Home?

A single [BrainGate Consortium](https://bciintel.com/companies/braingate) participant with [amyotrophic lateral sclerosis (ALS)](https://bciintel.com/glossary/als) has now logged more than 3,800 hours of independent, at-home use of an intracortical speech BCI — reaching 56 words per minute, 99% word accuracy on controlled testing with a 125,000-word vocabulary, and generating close to 2 million decoded words over nearly two years. Those figures, published in *Nature Medicine* and emerging from a UC Davis-led collaboration with Brown University and Mass General Brigham Neuroscience Institute, represent the most sustained real-world demonstration of unsupervised intracortical BCI operation ever reported in the published literature. The participant, Casey Harrell, a 47-year-old man with ALS-related tetraparesis and dysarthria, rated 92% of decoded sentences as accurate or mostly correct during daily use. Critically, he operated the system without any researcher present — resolving what has been the central translational bottleneck for intracortical communication BCIs: supervised-only deployment.

This is a Phase 1 feasibility study with a single participant. It is not a controlled trial, and the results cannot yet be generalized to the broader ALS or paralysis population.

---

## The Hardware and Implant Details

In 2023, UC Davis neurosurgeon and co-principal investigator David Brandman implanted four microelectrode arrays into Harrell's left precentral gyrus — the cortical region coordinating speech motor output. The arrays collectively record from 256 cortical electrodes, capturing single-unit and multi-unit activity that the decoding stack translates into two parallel output streams: a speech BCI for text generation and a movement BCI for [cursor control](https://bciintel.com/glossary/cursor-control).

The precentral gyrus targeting is noteworthy. Rather than targeting traditional language areas, this approach decodes the motor intention to speak — attempted articulation — from the primary motor cortex speech representation. That strategy has been consistent across the BrainGate speech BCI lineage and has proven robust even as ALS progressively reduces or eliminates overt speech output.

The [electrode array](https://bciintel.com/glossary/electrode-array) hardware itself is the same Utah Array-based architecture that has underpinned BrainGate research for over a decade. Signal longevity over nearly two years of daily use — described by co-senior author and neuroscientist Sergey Stavisky as "working very well for almost two years" — is clinically significant given that electrode performance degradation over months remains one of the unresolved challenges in intracortical BCI. The source does not provide granular signal quality metrics over time, so claims about long-term stability should be read as qualitative for now.

---

## What Changed From Prior BrainGate Speech BCI Work

Stavisky drew a direct comparison to the team's previous publication: in that earlier study, word decoding accuracy reached 97%, but the neuroprosthesis required a researcher on-site to set up each session. The current system eliminates that dependency entirely.

Three improvements are documented in the source:

1. **Accuracy**: Controlled testing moved from 97% to more than 99% word accuracy, with a 125,000-word vocabulary.
2. **Speed**: Harrell's average communication rate reached 56 words per minute — described as having increased over time since he began using the system.
3. **Independence**: Over 3,800 hours of near-daily home use without researcher support, with some sessions reportedly extending beyond 12 consecutive hours.

The clinical significance of the independence milestone is hard to overstate. Every prior high-performance intracortical BCI result — including earlier BrainGate work, as well as results from competing programs — has required trained personnel for setup, calibration, or session initiation. A system a participant can activate and use autonomously moves the technology from a research instrument toward something resembling an assistive device with real daily utility.

---

## The Dataset as a Research Asset

Lead author Nicholas Card, a postdoctoral scholar in UC Davis Neurological Surgery, and Stavisky both flagged the secondary scientific value of Harrell's participation: the 3,800 hours of home-use brain recordings, at single-neuron resolution, constitute what the team describes as the largest individual brain recording dataset with single-neuron resolution yet collected. That dataset will drive decoder improvements and deepen understanding of how the human motor cortex encodes speech over extended time periods and across disease progression — data that no animal model or short-session human recording study can replicate.

For the broader field, this matters. The ALS progression trajectory means Harrell's neural signals are evolving as the disease advances. A decoder that remains above 99% word accuracy under those conditions — if confirmed over longer follow-up — would provide a meaningful existence proof for adaptive decoding robustness.

---

## Skeptical Read: What This Study Cannot Tell Us

Several important caveats deserve direct attention:

**Single participant.** Every number reported — 56 WPM, 99% accuracy, 92% user-rated sentence accuracy — comes from one individual. ALS presents heterogeneously; Harrell's residual motor cortex function, the trajectory of his disease, and his specific implant placement may not generalize.

**No control arm.** There is no comparator group, no randomization, and no blinding. The 92% user-rated accuracy figure is self-reported by the participant, introducing potential response bias.

**Vocabulary-constrained testing.** The 99% figure applies to controlled testing conditions with a defined 125,000-word vocabulary. Real-world accuracy during unconstrained communication — names, technical terms, neologisms — likely differs, though the source does not provide that breakdown.

**Hardware longevity.** While the system "worked well" for nearly two years, the source does not report electrode yield or signal quality degradation curves. Intracortical arrays historically show declining channel counts over multi-year periods. The BrainGate2 trial is still enrolling, which means larger datasets across multiple participants will eventually allow better longevity characterization.

**Surgical access.** Four-array precentral gyrus implantation requires open craniotomy. That surgical burden limits the addressable patient population and will remain a central commercial barrier until fully wireless, minimally invasive intracortical solutions — or competitive endovascular approaches like those pursued by [Synchron](https://bciintel.com/companies/synchron) — mature further.

---

## Industry and Clinical Translation Implications

The publication arrives at an inflection point for the speech BCI sector. Neuralink has published human motor BCI data showing high-throughput cursor and typing control. Precision Neuroscience is advancing its Layer 7 cortical surface array through early human feasibility work. But none of those programs has yet reported multi-year, unsupervised home use data at the word-accuracy levels shown here.

For the intracortical BCI field specifically, this study shifts the conversation from "can it decode speech accurately in the lab?" — a question now answered affirmatively by multiple groups — to "can a patient use it daily, without a PhD student in the room?" That question has a partial affirmative answer for the first time, in a peer-reviewed publication.

The remaining translation gap is commercialization infrastructure: FDA regulatory pathway (BrainGate2 operates under an investigational device exemption), reimbursement, surgical training, and remote device management. None of those are addressed by this study. The BrainGate2 trial is still enrolling, and the source does not report a timeline for IDE expansion, De Novo submission, or PMA filing.

For teams working on motor cortex decoding applied to external effectors — robotic arms, powered wheelchairs, humanoid robot teleoperation — the decoder robustness and unsupervised operation findings from this study are directly relevant architecture signals. Coverage of those adjacent applications is tracked at [humanoidintel.ai](https://humanoidintel.ai).

---

## Key Takeaways

- **56 words per minute** average communication speed, **99% word accuracy** in controlled testing with a 125,000-word vocabulary — the strongest published combination of speed and accuracy for a home-use intracortical speech BCI.
- **3,800+ hours** of independent, researcher-free home use over nearly two years is a new benchmark for unsupervised intracortical BCI deployment.
- **Four microelectrode arrays, 256 cortical electrodes**, implanted in the left precentral gyrus in 2023 by UC Davis neurosurgeon David Brandman.
- **92% of sentences** rated accurate or mostly correct by the participant during daily real-world use.
- The study is a **single-participant Phase 1 feasibility result** within the BrainGate2 clinical trial — not a controlled trial — and results cannot be generalized without larger enrollment.
- The ~3,800 hours of single-neuron-resolution home recordings are described as the largest such individual dataset yet collected, with direct value for decoder development.
- BrainGate2 is **still enrolling participants**; no regulatory filing timeline is reported in the source.

---

## Frequently Asked Questions

**How fast can the BrainGate2 ALS participant communicate using the BCI?**
According to the *Nature Medicine* study, Casey Harrell averaged 56 words per minute using the UC Davis intracortical speech BCI — a rate that increased over time as he used the system daily at home.

**What is the word accuracy of the BrainGate2 speech BCI?**
The system achieved greater than 99% word accuracy in controlled testing using a 125,000-word vocabulary. During real-world daily use, Harrell rated 92% of decoded sentences as accurate or mostly correct.

**Can an ALS patient use a BCI at home without a researcher present?**
This study is the first peer-reviewed report of a person with ALS operating a high-performance intracortical speech BCI independently at home over an extended period — more than 3,800 hours across nearly two years. Prior BrainGate systems required on-site researcher support for each session.

**How many electrodes does the BrainGate2 speech BCI use?**
The implant consists of four microelectrode arrays recording from 256 cortical electrodes, placed in the left precentral gyrus — the region coordinating speech motor output.

**Is the BrainGate2 speech BCI approved by the FDA?**
No. BrainGate2 operates under an investigational device exemption (IDE) as a clinical trial. It is not commercially approved. The trial is still enrolling participants, and the source does not report a timeline for any regulatory submission toward approval.

**How does this BCI compare to other speech BCI programs?**
Multiple groups — including teams associated with UCSF, Stanford, and commercial programs — have demonstrated high-accuracy speech decoding in supervised research settings. The distinguishing result here is the combination of unsupervised home use, multi-year stability, and the volume of real-world decoded output (close to 2 million words). Direct accuracy comparisons across programs require caution given differing vocabulary sizes, task designs, and participant populations.